Charles Edmondson said:
Yep, the joys of computer extraction! It is always interesting when
there are questions on models, esp. some of the new BSIM models. I have
had several students say "We need a generic BSIMx.x model, can you help
us?" The problem is that, these models are not unique! So many of the
parameters interact in 'interesting' ways, that you can two models with
very, very different values, that simulate with exactly the same results!
As with all things with models, make sure it models the effects you
want. A generic Cap model will not model ESR or the failure modes of a
tantalum. Most opamp models don't model power supply drain. If a model
exhibits the behavior you need well enough, then use it!
Just so. A model is an abstraction that allows you to predict future
behavior. A good model simply predicts well enough to allow you
ignore what it doesn't predict.
Most device models are lower or higher order that real life devices,
which means that the equations will be under- or over-specified when
you try to match simulation to live measurements. In the
underspecified case, the simulation will only "match well" for a
handful of bias points. In the overspecified case (BSIM models
typically), the simulation will "match well" over many non-unique
parameter value sets - which is actually a more insidious situation.
The only way to get around either is to "know" the model and "know"
your simulation/design conditions so you can plan the extraction well.
As for "extracting a model from a datasheet", in most cases the
datasheet parameters are geared toward *hand-design* rather than
simulator design. For example the typical specs on a BJT are barely
sufficient to define even the basic Ebers-Moll BJT model, though you
can bootstrap the JFET model with typical JFET datasheet specs. If
you are doing transistor level design, you should be starting with a
hand-design anyway.